4 ms·

Previous discussion: https://news.ycombinator.com/item?id=9419784 https://news.ycombinator.com/item?id=9419784. I was really curious to see if there had been so

• tomp 8y ago
  Following peer review comments, Andreas has significantly revised his paper, and simplified some of the derivations, particularly in the "type inference" section, making the type derivation rules more algorithmic and easier to implement.
• tomp 8y ago
  Some further info. A follow-up paper 1ML with Special Effects [1] extends the original 1ML type system with effects polymorphism (i.e. that higher order functions such as ‘map’ can either be pure/applicative or impure/generative), but IMO at a huge loss of simplicity of the original paper - essentially you need a limited implementation of type classes! An important (in my opinion) extension of 1ML would be value tracking, which was implemented for modules in the earlier F-ing Modules paper, and would solve some of the remaining issues with ML module system. I tried doing that (only implementation, not theory, and without type inference), but it’s... hard (and I’m not sure it can be done keeping the simplicity of 1ML). [1] https://people.mpi-sws.org/~rossberg/1ml/1ml-effects.pdf https://people.mpi-sws.org/~rossberg/1ml/1ml-effects.pdf